OpenAI updates GPT-5.5 Instant behaviour in ChatGPT
OpenAI updated ChatGPT's default model to GPT-5.5 Instant in May 2026, replacing GPT-5.3 Instant for all users. Per OpenAI's launch blog, the update delivers 52.5% fewer hallucinated claims and 37.3% fewer inaccurate answers on challenging prompts, with responses that are tighter, less emoji-heavy, and more personalized using past chats and connected apps. A subsequent update removed the Canvas side-panel from both GPT-5.5 Instant and GPT-5.5 Thinking, moving writing and coding tasks into direct chat blocks. OpenAI also set retirement timelines for GPT-4.5 (June 27, 2026) and o3 (August 26, 2026) from ChatGPT. For teams building on the default model, the behavioral changes affect prompt reproducibility, response formatting, and evaluation baselines.
What changed
OpenAI updated ChatGPT's default model to GPT-5.5 Instant on May 5, 2026, replacing GPT-5.3 Instant for all users. The company describes the update as improving accuracy, response quality, and personalization for the hundreds of millions of users who rely on the model daily (OpenAI).
Accuracy improvements
In internal evaluations, GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts covering medicine, law, and finance, and reduced inaccurate claims by 37.3% on especially challenging conversations that users had flagged for factual errors (OpenAI). The update also improves performance on image and photo analysis, STEM questions, and decisions about when to trigger web search (OpenAI, 9to5Mac).
Response style
OpenAI says responses are "tighter and more to-the-point without losing substance, while keeping the warmth and personality that makes ChatGPT enjoyable to use." The update reduces verbosity and overformatting, cuts unnecessary follow-up questions, and removes gratuitous emojis (OpenAI, 9to5Mac). A subsequent update removed the Canvas side-panel feature from GPT-5.5 Instant and GPT-5.5 Thinking; writing and coding tasks now run directly in chat through specialized writing and code blocks, with Canvas remaining accessible through older legacy models for paying users during a transition period (The Decoder).
Model deprecations
Alongside the behavior update, OpenAI announced retirement dates for two older ChatGPT models: GPT-4.5 retired from ChatGPT on June 27, 2026 (30-day sunset), and o3 is set to leave ChatGPT on August 26, 2026 (90-day sunset). Both remain available in the API for now (The Decoder).
Personalization
The update introduces memory sources across all ChatGPT models. When a response is personalized using saved memories, past chats, or connected apps such as Gmail, ChatGPT now displays which context it used and lets users delete or correct it. GPT-5.5 Instant is more effective at drawing on past chats and files to tailor answers, reducing the need for users to repeat context (OpenAI).
For practitioners
The routing system allows GPT-5.5 Instant to automatically switch to GPT-5.5 Thinking for complex requests; users can configure thinking effort and toggle automatic switching. The help article describes three tiers -- Instant, Thinking, and Pro -- each targeting different latency and capability tradeoffs (OpenAI Help Center). Behavioral changes to the default model affect prompt-response dynamics, evaluation baselines, and monitoring pipelines for teams building on ChatGPT. The canvas removal and response-style shift from bullet-heavy formats to prose also alter output formatting in downstream applications.
Scoring Rationale
GPT-5.5 Instant is the default model for hundreds of millions of ChatGPT users, so accuracy gains and formatting changes matter at scale. However, this is a behavioral tuning and UX update rather than a new model release or capability tier, placing it in the solid-but-not-major range.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
